Create app.py

app.py

@@ -0,0 +1,86 @@
+import os
+import chess
+import numpy as np
+import onnxruntime as ort
+from fastapi import FastAPI, Request
+from fastapi.responses import HTMLResponse, JSONResponse
+from fastapi.staticfiles import StaticFiles
+from fastapi.templating import Jinja2Templates
+from huggingface_hub import hf_hub_download
+
+app = FastAPI()
+
+# ১. অ্যাসেটস ও টেম্পলেট মাউন্ট করা
+app.mount("/static", StaticFiles(directory="static"), name="static")
+templates = Jinja2Templates(directory="templates")
+
+# ২. মডেল ডাউনলোড ও লোড (Hugging Face থেকে সরাসরি)
+MODEL_REPO = "GambitFlow/Synapse-Edge"
+MODEL_FILENAME = "v1/synapse_edge_v1.onnx"
+
+print("📥 Loading Flagship Model...")
+model_path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILENAME)
+session = ort.InferenceSession(model_path)
+print("✅ Model Ready for Inference.")
+
+# ৩. ১১৯-চ্যানেল টেনসর লজিক (Inference-Optimized)
+def get_tensor(fen):
+    board = chess.Board(fen)
+    tensor = np.zeros((1, 119, 8, 8), dtype=np.float32)
+    # Piece Planes (12)
+    pm = {'P':0, 'N':1, 'B':2, 'R':3, 'Q':4, 'K':5, 'p':6, 'n':7, 'b':8, 'r':9, 'q':10, 'k':11}
+    for sq, pc in board.piece_map().items():
+        r, f = divmod(sq, 8)
+        tensor[0, pm[pc.symbol()], 7-r, f] = 1.0
+    # Attack Maps (12)
+    for c_idx, color in enumerate([chess.WHITE, chess.BLACK]):
+        for p_idx, pt in enumerate([chess.PAWN, chess.KNIGHT, chess.BISHOP, chess.ROOK, chess.QUEEN, chess.KING]):
+            mask = 0
+            for sq in board.pieces(pt, color): mask |= board.attacks(sq)
+            for sq in range(64):
+                if (mask >> sq) & 1:
+                    r, f = divmod(sq, 8)
+                    tensor[0, 12 + c_idx*6 + p_idx, 7-r, f] = 1.0
+    # Auxiliary (Castling & Turn)
+    idx = 33
+    if board.has_kingside_castling_rights(chess.WHITE): tensor[0, idx, :, :] = 1
+    if board.has_queenside_castling_rights(chess.WHITE): tensor[0, idx+1, :, :] = 1
+    if board.has_kingside_castling_rights(chess.BLACK): tensor[0, idx+2, :, :] = 1
+    if board.has_queenside_castling_rights(chess.BLACK): tensor[0, idx+3, :, :] = 1
+    if board.turn == chess.WHITE: tensor[0, 37, :, :] = 1
+    return tensor
+
+# ৪. এপিআই এন্ডপয়েন্ট (চাল পাওয়ার জন্য)
+@app.post("/get_move")
+async def get_move(data: dict):
+    fen = data.get("fen")
+    try:
+        board = chess.Board(fen)
+        tensor = get_tensor(fen)
+        
+        # Inference
+        policy, value, tactical, phase = session.run(None, {"input": tensor})
+        
+        # ১০০০% নির্ভুল UCI Move জেনারেশন (Simplified for now)
+        # ফিউচারে এখানে আমরা MCTS যোগ করব, আপাতত বেস্ট লিগ্যাল মুভ সর্টিং
+        legal_moves = list(board.legal_moves)
+        # এখানে আমরা পলিসি হেড থেকে মুভ সর্ট করব
+        best_move = str(legal_moves[0]) # Placeholder
+        
+        return JSONResponse({
+            "move": best_move,
+            "value": float(value[0][0]),
+            "tactical": float(tactical[0][0]),
+            "phase": int(np.argmax(phase[0]))
+        })
+    except Exception as e:
+        return JSONResponse({"error": str(e)}, status_code=400)
+
+# ৫. ফ্রন্টএন্ড রেন্ডারিং
+@app.get("/", response_class=HTMLResponse)
+async def read_root(request: Request):
+    return templates.TemplateResponse("index.html", {"request": request})
+
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run(app, host="0.0.0.0", port=7860)